Slashdot Mirror
IBM Wants CPU Time To Be A Metered Utility
kwertii writes "IBM CEO Samuel J. Palmisano announced a sweeping new business strategy yesterday, pledging $10,000,000,000 towards redefining computing as a metered utility. Corporate customers would have on-demand access to supercomputer-level resources, and would pay only for time actually used. The $10 billion is slated for acquisitions and research to put the supporting infrastructure in place. Will this model revolutionize the way companies compute, or is this plan doomed to be another PCjr?"
It took the rest of the computing world YEARS to match the color & sound of that baby. What, you don't remember CGA and speaker music? Tisk.
If my history books and gathered information is correct, that was a business model used in early computers. A company would lease CPU time to users, generally because the end users couldn't afford the massive costs involved to purchasing and maintaining them.
Now, I'm relatively young (mid 20's), but I recall people not even a half generation older than I telling stories about getting in trouble for running up large bills on their school's timeshare account.
I could see where this might be useful, but only for a small handful of customers. There are not very many users of supercomputer's out there right now. I can't see that number increasing much just by servicing new customers who could benefit from a supercomputer but couldn't otherwise justify it for a short term project.
If they are dumping 10 billion dollars into this, they must think they are going to get at least that much out of it. I seriously doubt that they could do so, not without ridicously overpricing their service. For small time users who don't need supercomputer levels, there are much cheaper ways to go. (Buy your own gear, lease your gear, etc)
I work for a specialized outsourcing outfit that manages storage for large customers (internet datacenters primarily). I know how much of a pain in the ass it is to accomplish what we do now. I could just see the mess people would get into by getting into a timeshare system like this.
A slip of the foot you may soon recover, but a slip of the tongue you may never get over. -Benjamin Franklin
In the long run, wouldn't it be cheaper just to hire a consulting firm to build you a cluster?
IBM does that, and they're not making the kind of money they'd like to at it recently. Neither is anybody else (e.g., EDS).
This appears to be IBM's bid to claim a larger share of a shrinking IT pie.
The SCO lawsuit makes me wish my company were in Utah. We need a new building.
"We view this as Palmisano's coming-out party," said Thomas Bittman, an analyst at Gartner Research. "The industry will be measuring IBM against this as a benchmark for years."
Well, here is Gartner Group, missing the boat again. SimUtility has been doing this for years now, but because IBM is getting in to the market its news?
Timesharing of computers is a very valid, and far from dead market for computing. There are a lot of companies that do not want to buy their own supercomputers, which will likely sit unused the majority of the time. As for the example of a car manufacturer doing testing on a new model, this already happens as do many other organizations.
- America's Cup boat designers
- Racing teams
- Natural Resource Explorers
- scientific organizations
- and many many more
We're not exactly talking about a new or even revived paradigm. Timesharing never died.
"The words of the prophets are written on the Slashdot walls."
"Irving Wladawksy-Berger, vice-president of technology and strategy for IBM's Server Group, is a 32-year IBM veteran whose career has included stints in research, product development, business management, and strategic planning. In 1995 he was handpicked by CEO Lou Gerstner to figure out how to make the Internet a core part of IBM's business. He is still on that mission, although his latest focus is on two next-generation technologies: grid computing and autonomic systems. Wladawksy-Berger believes the Internet is on the verge of becoming a global virtual computer, like a utility power grid, with computing resources available on demand."
Our article today sounds like batch:
"computing power of a supercomputer for a short period" although they do go on to say "Other services could be delivered in much the same way".
1. Transferring product from generator (IBM supercomputer) to location. If you've just used 1 month of supercomputer time to model DNA folding, how will IBM transfer that data back to you? What if the computations and use are faster than the transmission rate?
...especially if its Windows run and goes down once a week, cutting into your bought utility time.
Well, all that you would need at your location would be the equivalent of an Xterminal, and you would have all you need. Why would you need more than visualization of the data at your location? If it is a metered utility, you should be able to access it from anywhere negating the need for data transfer from their cluster of supercomputers.
I doubt that they would use a system that goes down. Often supercomputers are clustered and use a common set of storage space that would allow migration of users and processes between systems. There should be minimal downtime in the final system-- the equivalent of current utilities. Also, they would likely only go down when your other utilties went out (lines cut, etc).
What if IBM becomes the only utility and charges way more than it should - there's no competition so Company A can't shop around. Along this same vien, if Company A is smart enough, they'll never enter into a utility agreement with IBM if they can generate their own computing cycles and be sure that they'll always be there, versus putting all their eggs in one basket.
If IBM did this and was successful, I'd feel sure that Sun, MS, Intel, and maybe others (does Tera still exist?) would start their own shops as competition. And companies are already putting their eggs all in one basket, but now it's just a basket that is their IT department.
I can't see supercomputing cycles as being something that is commodity, or for that matter, something I (or any company) needs to buy on a metered basis.
So, as your desktop you have access to this system. Maybe you are using only 20 CPU minutes per month as a standard desktop user. Imagine a company that has 10k users that would only use 20 CPU minutes per month. I'd think it would make sense in that case. Similar systems already exist, and they're called ASP's (Application Service Providers), and they already work on a similar concept.
The DOD and others already sell supercomputer CPU hours. I had a friend who had ~100000 CPU hours available to him on ASCI Red (for rocket and combustion fluid dynamics simulations). IBM is just formalizing it a bit more.
> There is no "faceless" technician at the other end
There is no fire-able individual that gets a performance review from the company. If they're unhappy with the outsourced datacenter performance, they have only two recourses: cancel the contract or sue, and I assume that contract agreements would most likely preclude the latter. The human element is completely being overlooked in these equations. Managers like pulling their staff together into a conference room and whipping their butts in times of crisis, making them feel in control of the company. Outsourcing precludes that. Sure, it will be (and has been) tried anyway, and will (and pretty much has in the case of ASPs) fail. But be my guest.
> What consumes most of the bandwidth in an internal company network is actually "raw" data.
[...]
> This means that the only bandwidth being used by your company will be to display web pages.
Hmm? Database queries are actually quite network efficient and in many respects very similar to HTTP. You send a query and get back a recordset. If you used a thin client instead, most of the information inside the recordset would likely travel across the network anyway, only in the form of more bloated ASCII-inside-HTML (to be displayed say inside an HTML table). And if the web server and database server don't reside on the same machine, you'd actually DOUBLE the network traffic.
Many cases can be made for browser-based thin client computing, but reduced network traffic definitely isn't one of them. There's nothing network efficient about stateless gobs of ASCII and graphics.
Another thing is that, as you mentioned, the ASP model is mainly suitable for web applications. Unfortunatly, that is still not the majority of applications in many corporations. There are still no satisfactory web versions of office applications, and there probably never will, because they're intrinsically client-side; if you insist on serving them via a browser, they will still end up mostly executing code (ActiveX, Java, JavaScript etc.) on the client side, but inside a sandbox, adding much headache and little benefit (think saving and printing).
Disclaimer: I'm a member of the Blue Collective.
It's called Capacity Upgrade on Demand (CUoD). Check IBM's site for more info.
The machines generally have all their processor slots populated, and IBM can remotely toggle them on when you need more CPU power for a workload. This is a cost effective way of providing users with the ability to upgrade as their computing needs grow. The cost of the idle CPU's is marginal when the entire system cost is taken into account.
What's that? What prevents a user from toggling on an idle CPU themselves? Nothing...other than invalidating their multi-million dollar maintenance contract with IBM.
"I disapprove of what you say, but I will defend to the death your right to say it."
- Evelyn Beatrice Hall
IBM has been toying with this idea internally for quite a while now. I tell you, it's a bad idea. Internal service delivery organizations use this model as a method to milk other organizations of money. They publish reports accounting for every query and every application run. Managers and accountants pour over these reports and DISCOURAGE people from running queries or applications unless "absolutely necessary". We end up with multimillion dollar S/390 sysplex's that sit idle.
There is another fundamental fallacy here... that is, CPU cycles are not a precious commodity. How can you sell the air? There are plenty of ways to mass computing power these days. The Arabs can set the price of crude oil because, they are the only ones with inexpensive crude oil. These days, everyone has access to more "processor cycles" than they require.
Within IBM, organizations are forced to use these models by management. Other companies will not choose to place such limitations on themselves.